From c5083422cdbc3263af87da9b7a3b95b1c2125321 Mon Sep 17 00:00:00 2001
From: erman-gurses <erman@nod-labs.com>
Date: Wed, 5 Feb 2025 14:46:03 -0600
Subject: [PATCH] Debug on CI

Signed-off-by: erman-gurses <erman@nod-labs.com>
---
 .../LLVMGPU/LLVMGPUSelectLoweringStrategy.cpp |  2 +-
 .../compiler/Codegen/LLVMGPU/Verifiers.cpp    | 66 ++++++++++---------
 2 files changed, 35 insertions(+), 33 deletions(-)

diff --git a/compiler/src/iree/compiler/Codegen/LLVMGPU/LLVMGPUSelectLoweringStrategy.cpp b/compiler/src/iree/compiler/Codegen/LLVMGPU/LLVMGPUSelectLoweringStrategy.cpp
index e4632e64bfa0..5a18c3188c51 100644
--- a/compiler/src/iree/compiler/Codegen/LLVMGPU/LLVMGPUSelectLoweringStrategy.cpp
+++ b/compiler/src/iree/compiler/Codegen/LLVMGPU/LLVMGPUSelectLoweringStrategy.cpp
@@ -48,7 +48,7 @@ verifyLoweringConfiguration(FunctionOpInterface funcOp,
     auto codegenLoweringConfig =
         getLoweringConfig<IREE::Codegen::LoweringConfigAttr>(op);
     auto gpuLoweringConfig =
-        getLoweringConfig<IREE::GPU::LoweringConfigAttr>(op);
+        dyn_cast_or_null<IREE::GPU::LoweringConfigAttr>(getLoweringConfig(op));
     if (!codegenLoweringConfig && !gpuLoweringConfig)
       return WalkResult::advance();
     return verificationFn(op, codegenLoweringConfig, translationInfo,
diff --git a/compiler/src/iree/compiler/Codegen/LLVMGPU/Verifiers.cpp b/compiler/src/iree/compiler/Codegen/LLVMGPU/Verifiers.cpp
index 2d776dbeabd6..2dbbf3aac632 100644
--- a/compiler/src/iree/compiler/Codegen/LLVMGPU/Verifiers.cpp
+++ b/compiler/src/iree/compiler/Codegen/LLVMGPU/Verifiers.cpp
@@ -80,51 +80,53 @@ verifyGPUMatmulPipeline(Operation *op,
 
   if (pipeline != CodeGenPipeline::LLVMGPUMatmulTensorCore &&
       pipeline != CodeGenPipeline::LLVMGPUMatmulTensorCoreMmaSync &&
-      pipeline != CodeGenPipeline::LLVMGPUTileAndFuse &&
       pipeline != CodeGenPipeline::LLVMGPUVectorDistribute) {
     return success();
   }
 
+  // Only verify batched and unbatched matmul.
+  if (!isa<linalg::MatmulOp, linalg::BatchMatmulOp>(op)) {
+    return success();
+  }
+
   uint32_t reduction = static_cast<uint32_t>(IREE::GPU::TilingLevel::Reduction);
   uint numLoops = llvm::cast<linalg::LinalgOp>(op).getNumLoops();
-
+  size_t size = numLoops;
+  numLoops = reduction;
+  size = size + 0;
   SmallVector<int64_t> reductionTileSizes =
       gpuLoweringConfig.getStaticTilingLevelSizes(reduction, op);
-  size_t size = reductionTileSizes.size();
+
+  if (!reductionTileSizes.empty()) {
+    size = reductionTileSizes.size();
+  }
+
   // if (size > numLoops) {
   //   return op->emitOpError("expected number of reduction tile size is equal
-  //   or "
-  //                          "less than number of loops");
+  //   or less than number of loops");
+  // }
+  // for (size_t i = 0; i < size; ++i) {
+  //   if (reductionTileSizes[i] > 0 &&
+  //       llvm::cast<linalg::LinalgOp>(op).getIteratorTypesArray()[i] !=
+  //           utils::IteratorType::reduction) {
+  //     return op->emitOpError(
+  //         "expected to non-zero reduction tile has reduction iterator");
+  //   }
   // }
-  for (size_t i = 0; i < size; ++i) {
-    if (reductionTileSizes[i] > 0 &&
-        llvm::cast<linalg::LinalgOp>(op).getIteratorTypesArray()[i] !=
-            utils::IteratorType::reduction) {
-      return op->emitOpError(
-          "expected to non-zero reduction tile has reduction iterator");
-    }
-  }
-
-  SmallVector<int64_t> workgroupTileSizes =
-      gpuLoweringConfig.getWorkgroupTileSizes();
-  size = workgroupTileSizes.size();
-
-  for (size_t i = 0; i < size; ++i) {
-    if (workgroupTileSizes[i] > 0 &&
-        llvm::cast<linalg::LinalgOp>(op).getIteratorTypesArray()[i] !=
-            utils::IteratorType::parallel) {
-      return op->emitOpError(
-          "expected to non-zero workgroup tile has parallel iterator");
-    }
-  }
 
-  if (pipeline == CodeGenPipeline::LLVMGPUTileAndFuse ||
-      pipeline == CodeGenPipeline::LLVMGPUVectorDistribute) {
-    return success();
-  }
+  // SmallVector<int64_t> workgroupTileSizes =
+  //     gpuLoweringConfig.getWorkgroupTileSizes();
+  // size = workgroupTileSizes.size();
+  // if (size) {
+  //   for (size_t i = 0; i < size; ++i) {
+  //     if (workgroupTileSizes[i] > 0 &&
+  //         llvm::cast<linalg::LinalgOp>(op).getIteratorTypesArray()[i] !=
+  //             utils::IteratorType::parallel) {
+  //       return op->emitOpError(
+  //           "expected to non-zero workgroup tile has parallel iterator");
+  // }
 
-  // Only verify batched and unbatched matmul.
-  if (!isa<linalg::MatmulOp, linalg::BatchMatmulOp>(op)) {
+  if (pipeline == CodeGenPipeline::LLVMGPUVectorDistribute) {
     return success();
   }